Median Filtering Detection of Small-Size Image Using AlexCaps-Network

Abstract

Digital image forgers often use various image processing software to maliciously tamper with image contents, and then use some anti-forensics techniques such as median filtering to hide the obvious traces of these tampered images. Therefore, median filtering detection is one of the key technologies in the field of image forensics. Recently, with the rapid development of the deep learning, more and more researchers have proposed many image median filtering detection algorithms based on deep learning. Deep learning method can automatically extract the image median filtering features and unify them with classification steps in a deep learning model, which has better detection performance than traditional algorithms. However, existing methods based on deep learning still have the promotion space when facing small size or highly compressed images. To solve this problem, a median filtering detection method of small-size image using AlexCaps-network is proposed in this paper. AlexCaps-network is a joint network combining the classical network Alexnet and Capsule network. Firstly, in order to cope with the difficulty of extracting median filtering features caused by the low-resolution of small size and highly compressed image blocks, we add image preprocessing layer to the first layer of the network to enhance the trace of median filtering. Secondly, the general feature of median filtering in learning images is extracted by shallow ordinary convolutional neural network. The capsule network layer extracts the more complex spatial information in the median filtering image by dynamic routing algorithm and predicts the results. Finally, the experimental results show that the effective detection performance of our proposed method for small size and highly compressed images, even though the size is 16 × 16 image blocks and QF of compression is 70, is still good.